Inventory Optimization For 900 Stores
  • Instagram - White Circle
  • GFAIVE LinkedIn page
  • GFAIVE Facebook page
  • Twitter - White Circle
  • YouTube - White Circle

© 2020 by GFAIVE - ML demand forecasting for fashion retail. Privacy Policy

Inventory Optimization Across 900+ Stores In 4 Countries

Challenges and Industry

An international retail company selling a very wide range of fashion and apparel items (clothing, footwear and accessories), approached GFAIVE. The company has more than 900 stores in over 300 cities in Russia, Belarus, Kazakhstan, and Ukraine. With continuous expansion, their goal was to become a data driven company but with more than 1.6 million transactions/month and hundreds of item categories, they were having trouble analyzing all that data. One of their primary objectives was to reduce inventory levels and decrease holding costs, while maintaining a level of certainty that there would always be stock wherever and whenever it was needed.


The challenge for GFAIVE was to build a solution that would automatically process large amounts of data and would be used as a base and foundation for optimizing inventory decision making.

Solution

 

Our project started with a deep data analysis and hypothesis testing phase. The goal of the phase was to get an overall view of the company’s business processes from a data perspective and identify potential risks. First, we detected and analyzed abnormal points in the data and identified correlation patterns, seasonal factors and
trends. After that we analyzed sales (the factors that impact sales), out of stock situations and over stock situations (their frequency, reasons why they occur), illiquid stocks and other points.

 

After that we moved on to the predictive model development phase. Here we developed a highly accurate demand forecasting model that simultaneously generated weekly and monthly forecasts for each of the client’s item categories one season ahead. Different algorithms, from classical SARIMA to techniques like XGBoost , Prophet and Neural Networks, were tested. As a result, a hybrid combination of algorithms was applied.
 

The next phase was the optimization task. The goal was to minimize warehousing costs while maximizing sales by recommending the necessary amount of goods in each store and the insurance supply level. The optimization task considered factors like warehouse capacity, delivery routes, shipment time, store capacity, and other related factors which have significant impact on sales and warehousing costs.
 

The last phase included UI development (interactive, intuitive dashboards were designed and developed) and integration into clients’ existing daily inventory planning systems.

Results

 

1. The client was able to control inventory on a much more detailed level, they were now able to understand:

  • What product categories should be ordered and in what amounts

 

  • How much stock they needed to have at every given time and when to place orders

 

  • How to allocate inventory across different store locations

 

2. The implementation of our machine learning based solution into their inventory ordering decreased warehousing costs by 10%.

3. The company was able to improve overall customer satisfaction by reducing the
number of out of stock situations by over 50%.